AMD was seemingly preparing for a massive RDNA 4 GPU before dropping it. A shared document indicates that Team Red considered a graphics card that doubles the compute units of its current flagship, but has sadly dropped it like a hot rock.
According to a document shared by Kepler_L2 on Anandtech’s forums, AMD’s upcoming RDNA 4-based GPUs originally featured up to nine shader engines, further confirming previous rumours regarding a design combining nine chiplets. According to another user going by the name of Adroc_Thurston, high-end Radeon RX 8000 series GPUs could’ve boasted more than 200 compute units. This would’ve indicated a change in the SE/CU ratio compared to RDNA 3.
If AMD chucked GDDR6X or GDDR7 memory into the mix, it might’ve even put up a decent fight against Nvidia’s GeForce RTX 5000. Unfortunately, at least according to what we currently know, AMD no longer plans to release such GPUs. Instead, it’s focusing on mid-range Navi 4 chips using ageing GDDR6 VRAM.
For comparison, AMD’s current flagship, RX 7900 XTX, maxes out at 96 compute units. Now imagine double that performance with architectural and ray tracing improvements, and you got a beast. As usual, assuming all of this is true.
AMD could come out of its silence and announce that such chips were never cancelled and that its monster is surely coming our way. Until then we have to take what we got: a Navi 48 GPU with four shader engines with whatever CU ratio AMD decides to use.
No one knows why the Red Team would leave performance on the table. It could be manufacturing complications due to the high chiplet count or simply that AMD wants to conquer the mid/low-end as Intel is coming in hot with Battlemage.